Predicting the Unpredictable: Generative AI in Finance

The paper ‘StockGPT: A GenAI Model for Stock Prediction and Trading’ unveils StockGPT, an autoregressive model that learns and predicts the dynamics of daily U.S. stock returns. By treating stock return series as sequences, StockGPT leverages the power of generative AI to decode and anticipate stock market movements.

• StockGPT is trained directly on stock return history, considering each series as a token sequence.
• Utilizes the attention mechanism inherent to generative AI to uncover hidden representations predictive of future returns.
• In testing from 2001 to 2023, StockGPT’s predictions led to an annual return of 119% and a Sharpe ratio of 6.5 for a daily rebalanced portfolio.
• The model’s predictions override momentum and long-/short-term reversals, questioning the need for traditional price-based strategies.

This exploration into generative AI’s applications within the financial sector demonstrates its potential to surpass human capabilities in complex decision-making scenarios, like stock trading. StockGPT’s success heralds an exciting future where AI can offer valuable insights in realms traditionally dominated by human experts. The scalability of such technology can extend to areas such as automated code trading systems, continuing to push the boundaries of AI’s role in predictive analytics.